Selection Bias vs Cognitive Bias

"Evidence for evolution is good evidence against the existence of God."

"Well you would say that, but you're an atheist, so you're biased."

"We need to raise the minimum wage."

"Well you would say that, but you're a Democrat, so you're biased."

"I agree with Noam Chomsky's theories of linguistics."

"Well you would say that, but you're a huge Chomsky fan, so you're biased."

Each accusation of bias above seems to me flawed, since they get the causal arrows completely backwards. It's entirely plausible that someone becomes an atheist because of the evidence on evolution, that they join the Democratic party because they support a higher minimum wage, that they like Chomsky because of his linguistic theories. To turn around and claim their allegiances undermine the validity of their beliefs—when it's precisely these beliefs which led to their allegiances in the first place—seems entirely unfair.

Admittedly, there is a sort of bias here. Democrats are statistically biased towards wanting to raise the minimum wage, and the sort of person who thinks evolution is valid evidence against God is more likely to end up an atheist, and so on. But these are results of selection bias, not cognitive bias.

Selection bias occurs when the process by which we select or gather our data skews it, making it unrepresentative of the population at large: a Noam Chomsky fan is more likely to appreciate his work in linguistics than a random guy off the street.

But none of this implies any flaw in reasoning on the part of the Chomsky fan (or atheist or Democrat). An accusation of cognitive bias is much more serious, and does demand a reasoning error.

Granted, selection bias may be a good reason to take someone's opinion with a grain of salt; their opinion may not generalize to your own. If you're not already a Democrat, it's less likely their reasoning on the minimum wage will agree with yours. This is basically simple regression to the mean.

But people often act like this accusation of bias has more force than it really does. It's one thing to say, "If I thought about this longer, I'd probably come to a different conclusion than you." It's entirely different to say, "You're biased, so I can discount your argument."

Now one could make an argument for there being legitimate (cognitive) bias in the cases above. Perhaps the atheist, being committed to atheism, is motivated to misevaluate the evidence on evolution. But if that's what their opponent is claiming, then they should say so outright, and make a case for why they think so. They should clearly say what the evidence is, in what manner the atheist is misinterpreting it, and what the correct interpretation tells us.

Merely observing that they're an atheist does not make the case for bias in this sense, but the accuser acts like it does. The source of this confusion, I contend, is that being an atheist is a good case on its own for there being selection bias at play, and the accuser is conflating this with cognitive bias.

I most recently heard a real-life example of this from Neel Nanda, directed toward himself. In arguing for the importance of mechanistic interpretability to AI safety, he admitted that as a mech interp researcher, he's biased in evaluating its true merit. I think Neel was being needlessly hard on himself. Surely the primary reason that he works in this field is because he thinks it's important for AI safety. Why should him acting upon that belief be grounds to dismiss it?

(I've also had someone do the equivalent of the Chomsky example to me when I suggested that the Sequences were one of the most important things ever written. "You would say that, but you're an Eliezer Yudkowsky fan, so you're biased," ignoring that I'm a fan of him because I think his writing is so valuable.)

This all relates to a more general principle: when you accuse someone of bias, you should identify the actual mistakes they made. It's all too easy to find some bias; almost every argument has a little bit in there somewhere. But in pointing this out, you take on the responsibility to actually show what they did wrong before you start psychologizing at them. If the Democrat is wrong about the minimum wage, surely they're wrong for reasons independent of their party affiliation. State those reasons plainly, and then afterward you can try explaining how their partisanship may have clouded their thinking. And if you can't identify any mistake, maybe drop that whole "bias" thing.

(This also seems to me like another instance of the Argument From My Opponent Believes Something. In this case, it goes: Argument From My Opponent Believes Something That Aligns With Their Other Beliefs And Behaviors, Which Is Kinda Like Being Biased.)